Start with one decision people actually care about.
Choose a bounded use case: an AI community assistant, a learning tool, a content recommendation system, or a customer-support workflow. Avoid starting with a system that makes irreversible high-stakes decisions. Name one accountable organizer and give participants a clear way to raise concerns.
| When | Action | Evidence of completion |
|---|---|---|
| Days 1–7: define | Map affected groups. Recruit 20–30 participants with varied experiences. Record the system’s goal, limits, data use, and current failure rate. Hold one accessible listening session. | A one-page charter, baseline test set, and list of missing voices. |
| Days 8–14: deliberate | Collect proposals privately before showing totals. Discuss tradeoffs. Reserve space for minority concerns. Publish the organizer’s response to each shortlisted proposal. | Three candidate changes, their rationales, and a decision log. |
| Days 15–21: test | Implement one reversible change. Compare against the baseline using the same tasks. Review failure cases with people who did not design the change. | A before-and-after report with limitations, complaints, and subgroup observations. |
| Days 22–30: account | Pay agreed contributions if funds exist. Publish spending, participation gaps, and failures. Decide whether to continue, revise, or stop. | A public pilot report and a dated next decision. |
A small, funded DEO pilot
Illustrative budget: $1,000 total, with $600 for participant time, $200 for independent review, $100 for accessibility, and $100 contingency. This is a planning example, not a funded Edaptus offer. If no budget is available, state that plainly and limit the unpaid workload.
Give every participant a base payment for completing an agreed review, regardless of their opinion. If extra rewards recognize verified findings, publish the rubric and let contributors appeal. Keep financial rewards separate from voting power.
Measure what would change your mind
- Quality: errors or harmful outputs per 100 comparable test cases, with sample size.
- Voice: proportion of proposals receiving a reasoned response; whose views remain missing.
- Fairness: distribution of compensation and acceptance rates by contribution type.
- Agency: can participants explain, challenge, and reverse the decision?
- Learning: can people identify the main tradeoff after participating?
Small pilots provide directional evidence, not population-level proof. Predefine a pause trigger, such as a privacy leak, inability to withdraw, or a consequential failure without a working review path.
What you can do today
Individuals: take the quiz and write one specific AI concern. Builders: audit one agent’s permissions. Community organizers: host a 30-minute discussion. Employers: invite workers to propose how time savings should be shared. Funders: support independent evaluation and participant access.